Statements (45)
Predicate | Object |
---|---|
gptkbp:instanceOf |
distributed computing library
|
gptkbp:backend |
gptkb:UCC
gptkb:MPI gptkb:Gloo gptkb:NCCL |
gptkbp:developedBy |
gptkb:Meta_AI
|
gptkbp:documentation |
https://pytorch.org/docs/stable/distributed.html
|
gptkbp:hasComponent |
torch.distributed.launch
torch.distributed.algorithms torch.distributed.autograd torch.distributed.checkpoint torch.distributed.elastic torch.distributed.fsdp torch.distributed.optim torch.distributed.pipeline.sync torch.distributed.rpc torch.distributed.shard torch.distributed.tensor torchrun |
https://www.w3.org/2000/01/rdf-schema#label |
Torch Distributed
|
gptkbp:license |
gptkb:BSD
|
gptkbp:partOf |
gptkb:PyTorch
|
gptkbp:provides |
point-to-point communication
collective communication primitives process group abstraction |
gptkbp:supports |
gptkb:bar
gptkb:broadcaster gptkb:recv fault tolerance scatter remote procedure call data parallelism model parallelism multi-GPU training reduce send multi-node training all-reduce dynamic membership parameter server architecture elastic training gather |
gptkbp:usedFor |
distributed training
|
gptkbp:bfsParent |
gptkb:TorchElastic
|
gptkbp:bfsLayer |
7
|